A fully digital background calibration of timing skew in undersampling TI-ADC

Abstract

This paper proposes a fully digital calibration of timing mismatch for undersampling Time Interleaved Analog-to-Digital Converter (TI-ADC) employed in Software Defined Radio (SDR) receivers. The proposed calibration scheme employs an ideal differentiator filter, a Hilbert transform filter and a scaling factor to compute the derivative of the input in any Nyquist Band (NB). The efficiency of the proposed technique is shown using a four-channel undersampling 60 dB SNR TI-ADC clocked at 2.7 GHz. Monte Carlo simulations show SNDR and SFDR improvements of respectively, 18 dB and 21 dB over the first three NBs.

Hardware implementation of all digital calibration for undersampling TIADCs

“…[8] proposed an architecture to generate pseudo aliasing signals using Hadamard transform and a derivative operation. In [9], the authors extended Matsuno's work to generate the aliasing signals for bandpass signals as shown in Fig. 1.…”

“…Through the feedback loop, the estimated coefficients are updated every signal sample by an amount that is proportional to the product of compensated signal and the error signal [8], [9]. The proportional factor is the adaptive step of the LMS architecture.…”

“…In such architectures, the adaptive step is chosen as tradeoff between the convergence time and the precision of the estimated coefficient which affects the SNR/SFDR performance of the calibration. Especially, for calibration architecture that operates with input occupied at any NB, it is usefully hard to choose appropriate adaptive step [9]. In this paper, we propose a technique to improve the convergence speed as well as the estimation accuracy, hence, the SNR/SFDR performance.…”

“…Nevertheless, they required either an additional channel or a pilot input signal for calibration. Currently, a new blind calibration technique presented in [9] can overcome their aforementioned issues. However, it encountered the instability problem of the adaptive feedback loop with longer settling time and more oscillation on the convergence curves of estimate parameters.…”

“…However, it encountered the instability problem of the adaptive feedback loop with longer settling time and more oscillation on the convergence curves of estimate parameters. In this paper, a new LMS calibration architecture is proposed to improve the convergence speed as well as the SNDR/SFDR performance of the technique described in [9]. In addition, the hardware implementation of the proposed technique is designed and validated on FPGA platform.…”